US20190370979A1 - Feature point identification in sparse optical flow based tracking in a computer vision system - Google Patents

Abstract

A method for sparse optical flow based tracking in a computer vision system is provided that includes detecting feature points in a frame captured by a monocular camera in the computer vision system to generate a plurality of detected feature points, generating a binary image indicating locations of the detected feature points with a bit value of one, wherein all other locations in the binary image have a bit value of zero, generating another binary image indicating neighborhoods of currently tracked points, wherein locations of the neighborhoods in the binary image have a bit value of zero and all other locations in the binary image have a bit value of one, and performing a binary AND of the two binary images to generate another binary image, wherein locations in the binary image having a bit value of one indicate new feature points detected in the frame.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS
• This application is a continuation of prior U.S. application Ser. No. 15/266,149, filed Sep. 15, 2016, to Deepak Kumar Poddar, et al., which claims the benefit of India Provisional Patent Application No. 7079/CHE/2015 filed Dec. 30, 2015, each of which is incorporated herein by reference in its entirety.
BACKGROUND OF THE DISCLOSUREField of the Disclosure
• Embodiments of the present disclosure generally relate to a computer vision system, and more specifically relate to new feature point identification in sparse optical flow based tracking in a computer vision system.
Description of the Related Art
• A new class of embedded safety systems, referred to as advanced driver assistance systems (ADAS), has been introduced into vehicles to reduce human operation error. Such systems may provide functionality such as rear-view facing cameras, electronic stability control, collision warning, and vision-based pedestrian detection systems. Many of these systems use a monocular camera and rely on real time computer vision processing to detect and track objects in the field of view of the camera. Optical flow based tracking is a key component in computer vision processing such as, for example, structure from motion (SfM), object detection, ego motion, video compression, and video stabilization.
• One approach to optical flow based tracking that may be used in embedded safety systems is sparse optical flow based tracking. Sparse optical flow based tracking is a feature-based approach in which features, e.g., image edges, corners, etc., are identified and tracked across consecutive frames captured by a monocular camera. Given the real time processing requirements in embedded safety systems, performance improvements in aspects of sparse optical flow based tracking are desirable.
SUMMARY
• Embodiments of the present disclosure relate new feature point identification in sparse optical flow based tracking in a computer vision system. In one aspect, a method for sparse optical flow based tracking in a computer vision system is provided that includes detecting feature points in a frame captured by a monocular camera in the computer vision system to generate a plurality of detected feature points, generating a first binary image indicating locations of the detected feature points with a bit value of one, wherein all other locations in the first binary image have a bit value of zero, generating a second binary image indicating neighborhoods of currently tracked points, wherein locations of the neighborhoods in the second binary image have a bit value of zero and all other locations in the second binary image have a bit value of one, and performing a binary AND of the first binary image and the second binary image to generate a third binary image, wherein locations in the third binary image having a bit value of one indicate new feature points detected in the frame.
• In one aspect, a computer vision system is provided that includes a monocular camera configured to capture a two dimensional (2D) frame of a scene, a feature point detection component configured to detect a plurality of feature points in a frame, and a new feature point identification component configured to identify new feature points in the detected plurality of feature points by performing a binary AND of a first binary image and a second binary image to generate a third binary image, wherein locations in the first binary image having a bit value of one indicate locations of the detected plurality of feature points and all other locations in the first binary image have a bit value of zero, and wherein locations in the second binary image having a bit value of zero indicate neighborhoods of currently tracked feature points and all other locations in the second binary image have a bit value of one, and wherein locations of the third binary image having a bit value of one indicate new feature points.
• In one aspect, a computer readable medium storing software instructions that, when executed by one or more processors comprised in a computer vision system, cause the computer vision system to execute a method for sparse optical flow based tracking. The software instructions include instruction to cause detection of feature points in a frame captured by a monocular camera in the computer vision system to generate a plurality of detected feature points, generation of a first binary image indicating locations of the detected feature points with a bit value of one, wherein all other locations in the first binary image have a bit value of zero, generation of a second binary image indicating neighborhoods of currently tracked points, wherein locations of the neighborhoods in the second binary image have a bit value of zero and all other locations in the second binary image have a bit value of one, and performance of a binary AND of the first binary image and the second binary image to generate a third binary image, wherein locations in the third binary image having a bit value of one indicate new feature points detected in the frame.
BRIEF DESCRIPTION OF THE DRAWINGS
• Particular embodiments will now be described, by way of example only, and with reference to the accompanying drawings:
• FIG. 1 is a block diagram illustrating an example sparse optical flow based tracking system that may be used in a computer vision system;
• FIG. 2 is an example illustrating a track;
• FIG. 3 is an example illustrating new feature point identification;
• FIG. 4 andFIGS. 5A-5C are examples illustrating new feature point identification;
• FIG. 6 is a flow diagram of a method for new feature point detection in a computer vision system; and
• FIG. 7 is a high level block diagram of an example multiprocessor system-on-a-chip (SOC) configured for use in a monocular camera-based automotive safety application.
DETAILED DESCRIPTION OF EMBODIMENTS OF THE DISCLOSURE
• Specific embodiments of the disclosure will now be described in detail with reference to the accompanying figures. Like elements in the various figures are denoted by like reference numerals for consistency.
• As previously mentioned, one approach to optical flow based tracking that may be used in embedded safety systems is sparse optical flow based tracking. In this approach, features, which may also be referred to as interest points or key points or feature points, are identified and tracked as they move from frame to frames in consecutive frames captured by a monocular camera.
• FIG. 1 is a block diagram illustrating an example sparse optical flow basedtracking system100 that may be used in a computer vision system. At a high level, the sparse optical flow basedtracking system100 detects and tracks two dimensional (2D) locations of feature points across consecutive frames, storing the 2D locations in tracks corresponding to the feature points. Thus, a track is a sequence of 2D frame coordinates of a feature point over time.FIG. 2 is a simple example illustrating atrack200. In this example, a feature point was initially detected in frame (t-4) at coordinates (x₀,y₀), and was tracked across the two subsequent frames at coordinates (x₁,y₁₁) in frame (t-3) and at coordinates (x₂,y₂) in frame (t-1). Thus, thetrack200 corresponding to the feature point is [(x₀,y₀), (x₁,y₁), (x₂,y₂)].
• Referring again toFIG. 1, the sparse optical flow basedtracking system100 includes a featurepoint detection component102, a new featurepoint identification component104, an imagepyramid generation component106, a sparseoptical flow component108, and atrack management component110. The input to the system is a frame captured by a monocular camera and the output is a set of tracks corresponding to feature points.
• The featurepoint detection component102 is configured to detect feature points in a frame (t-1). Any suitable technique for feature point detection may be used. For example, the feature point detection may be based on Harris corner detection or the features from accelerated segment test (FAST) detection. Harris corner detection is described, for example, in C. Harris and M. Stephens, “A Combined Corner and Edge Detector,” Proceedings of Fourth Alvey Vision Conference, Manchester, UK, pp. 147-151, 1988. FAST is described, for example, in E. Rosten and T. Drummond, “Machine Learning for High Speed Corner Detection,” Proceedings of 9^thEuropean Conference on Computer Vision, Vol. 1, Graz, Austria, May 7-13, 2006, pp. 430-443.
• The new featurepoint identification component104 is coupled to the featurepoint detection component102 to receive the detected feature points and to thetrack management component110 to receive the most recent 2D locations for the currently tracked feature points, i.e., the 2D locations of the tracked feature points in frame (t-1). The new featurepoint identification component104 is configured to analyze the detected feature points to identify any new feature points in the detected feature points, i.e., to eliminate any detected feature points that are in close proximity of the last tracked location of a feature point currently being tracked. More specifically, the new featurepoint identification component104 is configured to identify a feature point detected by the featurepoint detection component102 as a new feature when the 2D coordinates of the detected feature point are not within a small neighborhood of the most recent 2D coordinates of any tracked feature point. The size and shape of the neighborhood may be any suitable size and shape and may be determined empirically. In some embodiments, the neighborhood may be a 3×3 or a 5×5 square of pixels. The neighborhood used for new feature point identification and the neighborhood used by sparse optical flow may or may not be the same.
• FIG. 3 is a simple example illustrating new feature point identification. This example is a continuation of the example ofFIG. 2. Thefeature points302 and304 are feature points detected in frame (t-1) by the featurepoint detection component102. Thepoint300 is the last tracked location of the feature point corresponding to thetrack200. The box around thepoint300 represents the neighborhood around thepoint300. The detectedfeature point302 lies within the neighborhood and will be ignored by the new featurepoint identification component104 while the detectedfeature point304 is not in the neighborhood and will be identified as a new feature point for the frame (t-1).
• New feature point identification is performed because any feature points detected by the featurepoint detection component102 that lie within the neighborhoods of already tracked feature points do not provide any additional information to algorithms that use the tracking information as a detected feature point in the neighborhood of a tracked feature point is likely to be the tracked feature point and does not need to be tracked separately. Computation cycles spent tracking such feature points are redundant and can be better utilized by components in the computer vision system. Operation of the new featurepoint identification component104 is explained in more detail herein in reference toFIG. 4 andFIGS. 5A-5C.
• Referring again toFIG. 1, the imagepyramid generation component106 is configured to generate an image pyramid for a frame (t). That is, the imagepyramid generation component106 is configured to perform a multi-resolution decomposition of a frame to generate a pyramid of images of differing resolutions. Any suitable number of images in the image pyramid and any suitable resolutions of the images may be used. For example, the image pyramid may be a set of images in which each image is down-sampled by a factor of two both horizontally and vertically with respect to the image of the next higher resolution in the pyramid. The image pyramid enables analysis of the scene captured in the original frame at different “sizes”, e.g., over different distances.
• The sparseoptical flow component108 is coupled to the imagepyramid generation component106 to receive image pyramids for the most recent frame (t) and the previous frame (t-1), to the new featurepoint identification component104 to receive the new feature points for the previous frame (t-1), and to thetrack management component110 to receive the most recent 2D locations for the currently tracked feature points, i.e., the 2D locations of the tracked feature points in frame (t-1). The sparseoptical flow component108 is configured to perform point correspondence between the most recent frame and the previous frame using an image pyramid based sparse optical flow algorithm. An example of such an algorithm is described in V. Tarasenko and D Park, “Detection and Tracking over Image Pyramids using Lucas and Kanade Algorithm,” International Journal of Applied Engineering Research, Vol. 11, No. 9, pp. 6117-6120, 2016.
• In general, sparse optical flow is applied to attempt to match the new feature points (t-1) and the currently tracked feature points, i.e., the most recent 2D locations from tracks (t-1), with 2D locations in the most recent frame (t). The output of the sparseoptical flow component108 is the new feature points (t-1) and currently tracked feature points along with the 2D locations of the corresponding matching points in the current frame (t). For those new feature points (t-1) and currently tracked feature points for which no match was found in frame (t), a null 2D location is indicated.
• Thetrack management component110 is coupled to the sparseoptical flow component108 to receive the matched points. Thetrack management component110 is configured to manage the tracks, i.e., to start new tracks, to extend existing tracks, and to delete tracks that are no longer valid. Thetrack management component110 is configured to start a new track for a new feature point when the sparseoptical flow component108 indicates that a match was found for the new feature point in the most recent frame. In this latter case, the new track includes the 2D coordinates of the feature point in frame (t-1) and the 2D coordinates of the matching point in frame (t).
• Thetrack management component110 is also configured to extend an existing track when the sparseoptical flow component108 indicates that a match was found in the most recent frame for the feature point corresponding to the track. In this latter case, the 2D coordinates of the matching point are added to the track. Thetrack management component110 is also configured to delete an existing track when the sparseoptical flow component108 does not find a match in the most recent frame for the feature point corresponding to the track. In addition, thetrack management component110 is configured to provide the most recent 2D locations for the currently tracked feature points, i.e., the 2D locations of the tracked feature points in frame (t-1) to the new featurepoint identification component104 and the sparseoptical flow component108.
• FIG. 4 is an example illustrating the operation of the new featurepoint identification component104. As previously described, the inputs to the new featurepoint identification component104 are the 2D locations of the currently tracked feature points402 and the feature points404 detected in frame (t-1). The new featurepoint identification component104 is configured to compare the detectedfeature points404 to the currently tracked feature points to identify any detectedfeature points404 that are new. As previously described, if a detected feature point is not within a small neighborhood of a tracked feature point, the new featurepoint identification component104 is configured to identify the detected feature point as a new feature point. The output of the new featurepoint identification component104 is the new feature points406 detected in the previous frame (t-1), and the final feature points408 for the frame (t-1) to be processed by the sparseoptical flow component108 are the new feature points406 and the tracked feature points402.
• In the prior art, one approach to identifying new feature points in the detected feature points is to do a point by point search in which each detected feature point is compared to each of the tracked feature points to determine whether or not the detected feature point is within the small neighborhood of any of the tracked feature points. This approach can be very computationally intensive if the number of detected feature points and the number of tracked feature points is large, i.e., the problem is of complexity O(M*N), where M is the number of detected feature points and N is the number of tracked feature points. For example, if N=9000 and M=3000 and one cycle is used for each comparison of two 2D points, the total cycle time required to identify the new feature points may be approximately 27 mega cycles, which may be unacceptable given the real time processing requirements of embedded safety systems.
• Embodiments of the disclosure provide for new feature point identification with less computational complexity. Instead of comparing each detected feature point to a neighborhood around each tracked feature point, the new featurepoint identification component104 is configured to generate two binary images of the same dimensions as the frame size. In some embodiments, one of the binary images, i.e., the detected feature point binary image, any bit locations that correspond to the location of a detected feature point are set to one and all other bit locations are set to zero. In the other binary image, i.e., the tracked feature point binary image, any bit locations that correspond to a neighborhood around a tracked feature point are set to zero and all other bit locations are set to one.
• The new featurepoint identification component104 is further configured to perform a binary AND operation between the respective bit locations of the two binary images to generate a new feature point binary image in which a one in a bit location indicates a new feature point. The new featurepoint identification component104 is further configured to generate the new feature points406 by outputting the 2D coordinates of each bit location in the new feature point binary image having a value of one.
• FIGS. 5A-5C are a simple example illustrating the operation of the new featurepoint identification component104 in some embodiments.FIG. 5A shows an example tracked feature pointsbinary image500 for the tracked feature points502. In this example, there are ten tracked feature points and the neighborhood around a tracked feature point is assumed to be 3×3 points.FIG. 5B shows the corresponding detected feature pointsbinary image504 for the detected feature points506. In this example, there are eight detected feature points.FIG. 5C shows the result of the binary AND operation between the tracked feature pointsbinary image500 and the detected feature pointsbinary image504, i.e., the newfeature points image508. The newfeature points image508 is then converted to a set ofcoordinates512 for the new feature points in which the set includes the coordinates of each location in the image containing a one. Note that three of the detectedfeature points506 are within the neighborhood of a trackedfeature point502 and are not included the new feature points512 of the final feature points510.
• In some embodiments, a direct memory access (DMA) controller may be programmed to accelerate the generation of the tracked feature points binary image. While feature detection is being performed by thefeature detection component102, the DMA controller may perform chained operations to write the zero bit values in the neighborhoods of a tracked feature points binary image that is prefilled with one bit values. When the feature point detection is complete, the detected feature points image can be generated by the new featurepoint identification component104. Once both binary images are generated, the binary AND operation can be performed. In some embodiments, the output of thefeature detection component102 may be the detected feature points image.
• In some embodiments, the binary AND comparisons may be implemented on a single-instruction-multiple-data (SIMD) processor to accelerate the computation of the new feature point binary image. Assuming an 8-way SIMD processor, eight bytes can undergo an AND operation in one cycle. Thus, for a one mega pixel image, the total cycle time to AND the binary images will be ⅛ mega cycles. In the prior art approach to identifying new feature points, the cycle time depends on the number of detected feature points and the number of tracked feature points. In these embodiments, the cycle time depends on the image resolution and the number of bytes supported by the SIMD AND instruction.
• FIG. 6 is a flow diagram of a method for new feature point detection in a computer vision system. Initially, feature points are detected600 in a frame captured by a monocular camera. Feature point detection is previously described herein. A detected feature point binary image is generated602 based on the detected feature points and a tracked feature point binary image is generated604 based on the currently tracked feature points. These binary images are previously described herein. A binary AND operation is performed606 between the detected feature point binary image and the tracked feature point binary image to generate the new feature point binary image. The new feature point binary image is previously described herein. Finally, the 2D coordinates of the new feature points in the new feature point binary image are extracted608 from the new feature point binary image and output for further processing.
• FIG. 7 is a high level block diagram of an example multiprocessor system-on-a-chip (SOC)700 configured for use in a monocular camera-based ADAS. In particular, theexample SOC700 is an embodiment of the TDA3X SOC available from Texas Instruments, Inc. A high level description of the components of theSOC700 is provided herein. More detailed descriptions of example components may be found in M. Mody, et al., “High Performance Front Camera ADAS Applications on TI's TDA3X Platform,” Proceedings of 2015IEEE 22^ndInternational Conference on High Performance Computing, Dec. 16-19, 2015, Bangalore, India, pp. 456-463, and “TDA3x SOC Processors for Advanced Driver Assist Systems (ADAS) Technical Brief,” Texas Instruments, SPRT704A, October, 2014, pp. 1-6, which are incorporated by reference herein.
• TheSOC700 includes dual general purpose processors (GPP)702, dual digital signal processors (DSP)704, and avision processor706 coupled via ahigh speed interconnect722. TheSOC700 further includes a direct memory access (DMA)controller708, acamera capture component710 coupled to amonocular camera724, adisplay management component714, on-chip random access (RAM)memory716, e.g., a computer readable medium, and various input/output (I/O)peripherals720 all coupled to the processors via theinterconnect722. In addition, theSOC700 includes asafety component718 that includes safety related functionality to enable compliance with automotive safety requirements. Such functionality may include support for CRC (cyclic redundancy check) of data, clock comparator for drift detection, error signaling, windowed watch-dog timer, and self testing of the SOC for damage and failures. Software implementing sparse optical flow based tracking as described herein in which frames captured by themonocular camera724 are used may be stored in thememory716 and may execute on one or more programmable processors of theSOC700. In some embodiments, theDMA controller708 may be programmed to accelerate generation of the tracked feature points binary image as previously described herein.
Other Embodiments
• While the disclosure has been described with respect to a limited number of embodiments, those skilled in the art, having benefit of this disclosure, will appreciate that other embodiments can be devised which do not depart from the scope of the disclosure as disclosed herein.
• For example, embodiments have been described herein in which the spare optical flow based tracking used an image pyramid based sparse optical flow algorithm. One of ordinary skill in the art will understand embodiments in which other suitable sparse optical flow algorithms are used.
• In another example, embodiments have been described herein in which the sparse optical flow based tracking may be implemented as software instructions executed on processors in a multiprocessor SOC. One of ordinary skill in the art will understand that the sparse optical flow based tracking may be implemented as any suitable combination of software, firmware, and/or hardware. For example, some of the functionality may be implemented in one or more hardware accelerators, application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), etc.
• In another example, embodiments have been described herein in reference to automotive safety systems. One of ordinary skill in the art will understand embodiments for other computer vision applications, such as, for example, industrial applications, robotics, and consumer applications such as vacuum cleaners.
• Software instructions implementing all or portions of methods described herein may be initially stored in a computer-readable medium and loaded and executed by one or more processors. In some cases, the software instructions may be distributed via removable computer readable media, via a transmission path from computer readable media on another digital system, etc. Examples of computer-readable media include non-writable storage media such as read-only memory devices, writable storage media such as disks, flash memory, memory, or a combination thereof.
• Although method steps may be presented and described herein in a sequential fashion, one or more of the steps shown in the figures and described herein may be performed concurrently, may be combined, and/or may be performed in a different order than the order shown in the figures and/or described herein. Accordingly, embodiments should not be considered limited to the specific ordering of steps shown in the figures and/or described herein.
• Certain terms are used throughout the description and the claims to refer to particular system components. As one skilled in the art will appreciate, components in systems may be referred to by different names and/or may be combined in ways not shown herein without departing from the described functionality. This document does not intend to distinguish between components that differ in name but not function. In the description and in the claims, the terms “including” and “comprising” are used in an open-ended fashion, and thus should be interpreted to mean “including, but not limited to . . . .” Also, the term “couple” and derivatives thereof are intended to mean an indirect, direct, optical, and/or wireless electrical connection. Thus, if a first device couples to a second device, that connection may be through a direct electrical connection, through an indirect electrical connection via other devices and connections, through an optical electrical connection, and/or through a wireless electrical connection, for example.
• It is therefore contemplated that the appended claims will cover any such modifications of the embodiments as fall within the true scope of the disclosure.

Claims (20)

What is claimed is:

1. A method comprising:

receiving a frame of an image;

detecting a set of feature points within the frame, wherein the set of feature points includes a set of tracked feature points;

generating a first image that indicates locations of the set of feature points within the frame;

generating a second image that indicates neighborhoods of the set of tracked feature points; and

generating a third image based on the first image and the second image that indicates locations of a subset of the set of feature points that is different from the set of tracked feature points.

2. The method ofclaim 1, wherein the generating of the third image is such that the third image indicates a location of a first feature of the subset based on the first feature being outside the neighborhoods indicated by the second image.

3. The method ofclaim 1, wherein the frame is a first frame, and the set of tracked feature points is associated with a second frame that is prior to the first frame.

4. The method ofclaim 1 further comprising adding the subset of the set of feature points to the set of tracked feature points.

5. The method ofclaim 1, wherein each of the neighborhoods indicated by the second image is larger in area than a respective feature of the set of feature points.

6. The method ofclaim 1 further comprising determining a set of coordinates for each feature point of the subset of the set of feature points based on the third image.

7. The method ofclaim 1, wherein:

the generating of the first image is such that the locations of the set of feature points are indicated with a binary one and a remainder of the first image is indicated with a binary zero;

the generating of the second image is such that the neighborhoods of the set of tracked feature points are indicated with a binary zero and a remainder of the second image is indicated with a binary one; and

the generating of the third image includes applying an AND function to the first image and the second image.

8. A computer vision system comprising:

a camera configured to capture a frame;

a detection circuit coupled to the camera configured to detect feature points in the frame, wherein the feature points include a set of tracked feature points; and

a feature point identification circuit coupled to the detection circuit and configured to:

receive the detected feature points from the detection circuit;

receive the set of tracked feature points;

generate a first image that indicates locations of the detected feature points;

generate a second image that indicates neighborhoods of the set of tracked feature points; and

generate a third image based on the first image and the second image that indicates locations of a subset of the detected feature points that is outside the neighborhoods of the set of tracked feature points.

9. The computer vision system ofclaim 8, wherein the feature point identification circuit is further configured to add the subset of the detected feature points that is outside the neighborhoods to the set of tracked feature points.

10. The computer vision system ofclaim 9 further comprising a sparse optical flow circuit coupled to the feature point identification circuit to receive the set of tracked feature points that includes the added subset of the detected feature points.

11. The computer vision system ofclaim 8, wherein the feature point identification circuit is further configured to determine coordinates of the subset of detected feature points.

12. The computer vision system ofclaim 8, wherein each of the neighborhoods indicated by the second image is larger in area than a corresponding point of the set of tracked feature points.

13. The computer vision system ofclaim 12, wherein each of the neighborhoods has an area selected from a group consisting of: a 3×3 pixel area and a 5×5 pixel area.

14. The computer vision system ofclaim 8, wherein:

the first image indicates the locations of the detected feature points with a binary one and indicates a remainder of the first image with a binary zero;

the second image indicates the neighborhoods of the set of tracked feature points with a binary zero and indicates a remainder of the second image with a binary one; and

the feature point identification circuit is configured to generate the third image by performing an AND function on the first image and the second image.

15. A non-transitory computer readable medium storing software instructions that, when executed by one or more processors, cause the one or more processors to:

detect a set of feature points within a frame;

generate a first image that indicates locations of the set of feature points within the frame;

generate a second image that indicates neighborhoods of a first subset of the set of feature points; and

generate a third image based on the first image and the second image that indicates locations of a second subset of the set of feature points that is different from the first subset.

16. The non-transitory computer readable medium ofclaim 15, wherein the instructions that generate the third image are configured such that the second subset of the set of feature points is outside the neighborhoods of the first subset in the second image.

17. The non-transitory computer readable medium ofclaim 15 comprising further instructions that cause the one or more processors to determine a set of coordinates for each feature point of the second subset.

18. The non-transitory computer readable medium ofclaim 15, wherein the first subset of the set of feature points is a set of tracked feature points.

19. The non-transitory computer readable medium ofclaim 18 comprising further instructions that cause the one or more processors to add the second subset to the set of tracked feature points.

20. The non-transitory computer readable medium ofclaim 18, wherein the frame is a first frame, and the set of tracked feature points is associated with a second frame that is prior to the first frame.

Images